Learn
Estimate patterns and probability structure from bounded training evidence.
SRAI Book 6 · Chapter 1 · PU-B06-C01
Foundations of Generative AI
Understand how generative systems learn distributions, construct candidate outputs and require evidence, boundary, review and authority before consequential use.
01 / LEARNING OUTCOMES
Understand generation as probabilistic construction under accountable human authority.
- Explain how generative systems learn distributions and construct candidate outputs.
- Distinguish autoregressive, variational, adversarial and diffusion approaches.
- Relate training, inference, adaptation and sampling controls to system behaviour.
- Evaluate fidelity, diversity, utility, groundedness, robustness and safety.
- Identify provenance, privacy, security, fairness and uncertainty boundaries.
- Apply SRAI review and authorization principles in Health and Habitat.
02 / GENERATIVE SYSTEM CHAIN
Learn, condition, sample and review.
Condition
Frame the task through prompts, context, constraints and intended use.
Sample
Construct candidate outputs whose diversity depends on the model and decoding controls.
Review
Test evidence, provenance, boundaries and authority before consequential use.
03 / CONTROLLED LABORATORY
A generated output is a candidate construction—not an established fact.
The canonical notebook uses synthetic data to expose conditional generation, sampling behaviour, evaluation measures and the effects of control parameters.
Health and Habitat examples demonstrate why technical plausibility must remain separate from clinical, engineering, legal or operational authorization.
VERIFIED PRODUCTION UNIT
| Website documents | 5 PDF |
| Canonical notebook | 1 IPYNB |
| Manifest entries | 26 |
| Checksum entries | 27 |
| Data classification | Synthetic |
| Validation status | PASS |
Generation produces a probabilistic candidate.
Fluency, realism or confidence does not independently establish truth, suitability or permission to act.
People remain accountable for consequential use.
Evidence, boundary, review and explicit authority determine whether a candidate may support a decision.
04 / SRAI GENERATIVE-AI STANDARD
No generated output becomes authoritative merely because it is plausible.
- The information boundary and intended use are explicit.
- Grounding evidence and provenance remain inspectable.
- Uncertainty, failure modes and excluded uses are documented.
- Privacy, security, fairness and multilingual access are assessed.
- Human review is proportional to the consequences of error.
- Operational authority remains with an identified accountable role.
CONTROLLED RESOURCES
Read, reproduce, practise and review.
READ
View the complete GitHub production unit ↗
Audited Chapter 1
Controlled publication edition.
PDF ↗ REPRODUCECanonical notebook
Executable and independently verified.
IPYNB ↗ PRACTISEStudent assessment
Progress toward independent application.
PDF ↗ REVIEWInstructor solutions
Review the explained answers and assessment guidance.
PDF ↗ PRESENTLesson presentation
Review the complete visual presentation in controlled PDF format.
PDF ↗ RUNOpen in Google Colab
Run the approved notebook and examine controlled generative behaviour.
COLAB ↗ RELEASEView controlled GitHub release
Inspect the lesson, notebook, assessment, presentation and validation records.
GITHUB ↗ APPLYExecutive brief
Connect analytical controls to decisions.
PDF ↗